Introduction

DSH extends the Web UI via plugins. If you wish to perform microphone dictation in the DSH Web UI without a server, API key, or model download—just within the browser—you can use dsh-mic-input.

It is a speech input plugin for the DSH Web GUI input box: it fills the input box with real-time recognition results from the browser’s built-in Web Speech API, allowing you to continue editing or submitting messages.

What is this

dsh-mic-input is maintained by QT-Chen under the MIT license. It addresses the speech input problem in DSH Web UI input scenarios: it adds a microphone button to the input box toolbar. Clicking it starts dictation, and recognized text is entered into the input box.

The plugin runs on the browser side, with no server-side components, API keys, or model downloads.

Core Features

Microphone Button and Real-time Dictation

After installation, a microphone button will be added to the DSH Web GUI input box toolbar.

Click the button to start dictation; recognized text will be filled into the input box in real-time.

Draft Reconstruction and Deduplication

The plugin reconstructs the draft based on the text already present in the input box when recording starts, and deduplicates interim results to avoid duplicate text.

Silent Resume

If the browser automatically ends dictation after a brief pause, the plugin attempts to silently resume listening and preserve the session, supporting up to 6 times.

Punctuation and Language

Smart punctuation is enabled by default, supporting three modes: auto-completion, keep original, and remove.

Language selection includes:
- Follow system
- Chinese zh-CN
- English en-US

Auto-send and Settings Row

The plugin supports optional auto-send: automatically submitting the message after dictation ends.

Related settings are located at:
Settings -> General -> Speech Input

Permission denials or network errors are displayed in the button tooltip, with bilingual hints.

Installation and Enablement

First, install the plugin. Install from GitHub:

dsh plugin --profile web add github:QT-Chen/dsh-mic-input

Or install from npm:

dsh plugin --profile web add dsh-mic-input

After installation, restart dsh web or refresh the page. The plugin requires Node >=20.0.0; react >=18.2.0 is an optional peerDependency.

When installing from GitHub source code, pnpm ≥ 10 may require one-time build approval (allowBuilds); npm install does not have this step.

Typical Usage

  1. After installation, enter the DSH Web UI and click the microphone button on the right side of the input box.

  2. On first use, allow browser microphone permissions. 127.0.0.1 / localhost belong to a secure context, so the microphone can be used.

  3. Click the button to start recording; click the button again to stop and keep the transcript. The message is only automatically submitted after dictation ends if auto-send is enabled.

  4. If garbled text appears, check the language settings to ensure they match the language you are speaking.

  5. Recognition quality is limited by the browser’s built-in engine; Microsoft Edge (Microsoft Azure speech service) and Google Chrome (Google/Chrome speech service) may differ significantly.

Applicable Scenarios and Notes

Suitable for local browser usage of DSH Web UI for users who wish to input messages via voice.

The plugin has no server-side components, API keys, or model downloads; audio is handled by the browser’s speech service, and only the result text enters DSH. The plugin does not write to disk and has no storage.

Suggestions before and after use:
- Check the source code and MIT license before installation; the plugin runs with the permissions of the current dsh process.
- Confirm browser microphone permissions on first use.
- If recognition quality is not ideal, first check if the language setting matches the actual spoken language.
- If installing from GitHub source code, note that pnpm ≥ 10 may require a one-time build approval; npm install does not have this step.

Related Links

  • GitHub: https://github.com/QT-Chen/dsh-mic-input
  • Directory page (plugin clue address, not verified in the current collected data): https://www.skillhub.cn/plugins/QT-Chen/dsh-mic-input

The community directory is an independent site and is not equivalent to the DeepSeek official app store.